[Beowulf] UPS system for Linux cluster

[Mark Hahn]

> Actually, this isn't particularly surprising to folks who deal with
>electrical power. It *is* surprising to folks who see volts*amps (VA) and
>think it means watts.

hmm, I very rarely see any compute hardware which is has PF < .95.

> Most power conversion equipment has a higher VA rating than Watt rating
>(and, so do most loads, so it works out ok in practice)  PF (Power Factor) =
>Watts/VA, and a typical PF is around 0.85.

my impression is that .85 would have been typical for pre-PFC PSUs
(pretty common 5+ years ago).  or perhaps with lots of motors?

>can lead to nuisance trips of things like GFCIs, and if you plug in a whole
>rack full of power supplies at once, it can be pretty exciting (hmm, 20
>power supplies, each drawing 50A..)

I noticed the spec on a recent HP server was 30A inrush (and they didn't 
parameterize it by supply voltage).  so a rack (up to 40, say) * 30A 
does seem like a lot.

what actually happens if you fail to meet the inrush current?  I'd kinda
expect nothing bad to happen - how fast would breakers respond?  isn't 
the inrush really a design parameter - vendors could limit it to whatever
they wanted (such as 20% over nominal).

I did also note that servers are shipping with PSU's quoted at ~92%
efficiency, which is nice to see.  wonder if any of them will ever 
stick a little battery in the PSU like google...